Name		Name	Last commit message	Last commit date
parent directory ..
1.png		1.png
2.png		2.png
Dataset_Feature_Metadata.pdf		Dataset_Feature_Metadata.pdf
Prompt_Development(Different_Versions).pdf		Prompt_Development(Different_Versions).pdf
Prompt_with_Response.pdf		Prompt_with_Response.pdf
README.md		README.md
analytical_task_classification.png		analytical_task_classification.png

README.md

Generating Analytic Specifications for Data Visualization from Natural Language Queries using Large Language Models

1st, and 2nd prompt was our initial test of the capability of LLM with ChatGPT to see if it could generate Vega-Lite specifications given a subset of the dataset and a query.
Analytical Task Classification contains the prompt where we provided ChatGPT with a context of low-level analytic tasks using JSON objects describing analytic tasks, along with a subset of the dataset and a few natural language queries.
We also tested the capability of LLM using ChatGPT to extract metadata using our prompts and its ability to understand JSON objects describing analytic tasks, which is in Dataset Feature Metadata.
Prompt Development (Different Versions) contains the different versions of the prompt iteratively developed.
Prompt with Response contains the prompt with a subset of the dataset and a few natural language queries, with responses generated by ChatGPT detecting tasks, attributes detected, and the Vega-Lite specification for visualization.